Research On Image Classification Based On Sparse Representation

With more and more rapid development of multimedia and network technology, imageclassification has been widely used in many areas such as data mining, pattern recognition and soon. Achieving fast and accurate image classification has become a hot topic which people havebeen studying on over the years.The paper chooses the SVM and the weighted KNN method for image classification, thendeeply studies and proposes the image classification method based on sparse representation. Theexperimental results show that the SVM and the weighted KNN image classification methodsbased on sparse representation are reasonable and effective. This research work includes thefollowing aspects:1.The experiment has researched sparse representation Lasso algorithm, then makes use ofthe reconfigurable design of Lasso’s categorized characteristics to achieve a sparserepresentation subsystem. This system completes operation, solving sparse coefficients andsparse reconstruction. It makes optimized operation to the classification characteristics, in orderto improve the classification accuracy.2.A image classifier using SVM classification method is designed. The SVM imageclassifier is designed by support vector machine (Support Vector Machine, SVM). During theexperiment we classify the original features and the characteristics of sparse reconstruction. Theexperimental results show that combination of the sparse representation and SVM classificationmethod can be carried out more effectively to improve the accuracy.3.A image classifier using the weighed K-nearest neighbor classification algorithm isdesigned. The KNN image classifier is designed by K-nearest neighbor method (K-NearestNeighbor, KNN).During the experiment we classify the original features and the characteristicsof sparse reconstruction. The experimental results show that combination of the sparserepresentation and weighed KNN image classification method can be carried out moreeffectively to improve the classification accuracy. Furthermore, the accuracy with weighed KNNmethod is apparently higher than the KNN method accuracy.4.The image classification prototype system is designed. It uses the design ideas ofobject-oriented to design the system based on sparse representation. It validates the usefulnessand usability of the proposed method and the design method.